Research And Implementation Of Camera Track’s Self-adaptive Algorithm Based On Vision Navigation

In recent years, as computer vision-related technology and theory improving and developing, computer vision system is widely applied to visual inspection, visual navigation and automatic assembly field. In the vision system, since its price is low and it is access to rich information, camera is used as a tool of shooting, and processes the shot objects. However, as a passive sensor, if the environment becomes complex and severe, which is not suitable for human to manipulate camera, the camera is unable to achieve the desired results. In order to expand the scope of the camera, to achieve its autonomy and automation, the camera should be provided with the function of independent return and autonomous navigation. Especially in the unfamiliar and without stated goals environment, this performance becomes more important. One of the key tasks to realize camera’s autonomous return is to restore the camera’s trajectory.When the camera is moving in three dimensional space, if you want to achieve the self-return function, first of all need to do would be to recover the camera motion without GPS navigation and guidance information from ground. Therefore, it is a major work that how to restore the camera’s trajectory according to the obtained information. This paper is aimed at studying the global motion caused by the camera’s motion, analizing the global motion information contained by the image, obtaining the camera motion information, thus, achieving the description of the camera trajectory, in case of non-established targets and using only sport monocular camera to obtain images series.In this paper, the research is based on the relationship between the motion of each point in the image and the motion of the camera, Firstly, extract and match the feature points using the SIFT algorithm that is invariant to extract and match the feature points, by selecting the points of accurate positioning and having been detected and described as matching primitives during the research on the points’motion in the image. Then mismatching points should be eliminated by adopting approaches such as reverse matching, feature points on the slope angle, ect. Secondly, according to the obtained feature points set, calculate the parameters of global motion using the robust RANSAC model parameter estimation method, and the RANSAC method is improved to increase the efficiency of the algorithm. Finally, apply six parameters affine transformation model to describing the global motion, making the balance of efficiency and accuracy. Through the above steps the camera motion can be recovered. Because the number of the feature points extracted by SIFT algorithm is too many, it takes a lot of time to extract and match feature points, which affects the overall efficiency, the deep study on the extracting and matching of feature points is done. Setting the area-of-interest and reducing the sample rate can effectively reduce the number of feature points. Based on predicting the position area of matching points, the matching algorithm of feature points can effectively reduce the matching time.While ensuring the accuracy of the algorithm it improves overall efficiency, it is more suitable for real-time systems.